Shashwat Jha

I am an AI Engineer at Altimetrik and a New Graduate with MS in Computer Science and Engineering from the University of California, Riverside, where I was a Student Researcher under Prof. Amit K. Roy-Chowdhury.

Previously, I worked as a Data Scientist for a year at Axio (Acquired by Amazon), a leading digital lender in India backed by the likes of Peak XV (Sequoia Capital) and more. Before that, I had spent 6 months as an Intern at Bain & Company in the BCN division, contributing to AI-driven solution for global clients which led to a Full-time Offer

I was also a Student Research Scholar at the Artificial Intelligence Institute – University of South Carolina (AIISC), where I worked with Dr. Amit Sheth, Dr. Amitava Das, and Aman Chadha.

I completed my undergraduate studies at Birla Institute of Technology, Mesra, where I worked under the guidance of Dr. Raju Poddar and Dr. Vijay Nath on Deep learning applications across domains leading to multiple international conference publications, and was also a recipient of G.P Birla Scholarship for REU program at Northeastern University

Experience

A summary of my professional journey and research experiences

ASSOCIATE AI ENGINEER

ALTIMETRIK

June 2025 - Present

Southfield, MI

Full-time: Summer/Winter breaks and Post Graduation; Part-time: Academic Terms

Developed Full-Stack Agentic AI Database system using LangGraph for over 5M+ records across Data Warehouse with Next.js front-end and Knowledge Graph reinforced response routing, resulting in 8% improvement across retrieval metrics
Implemented Redis-backed session-aware ontology caching system using ElastiCache and AWS Lambda reducing query response time by 3x while achieving context recall and relevancy of 0.99 across multi-session Agentic AI systems
Automated CI/CD pipelines with Github Actions and set up observability framework using Langsmith and Ragas for monitoring and evaluation

STUDENT RESEARCHER

BOURNS COLLEGE OF ENGINEERING - Vision and Learning Group

January 2025 - March 2026

Riverside, CA

Researching Deepfake detection under network induced corruptions over live communication channels in production environments under Prof. Amit K. Roy-Chowdhury, developing novel methods to ensure accurate detection of deepfakes to enhance model reliability under real-world scenarios.
Advancing Safety Alignment in Vision-Language Models, designing robust defenses against adversarial attacks embedded in both images and textual inputs, ensuring secure and reliable AI performance.
Collaborating with the Collaborative Intelligence Systems Lab under Prof. Hang Qiu to develop a benchmark dataset for multi-agent perception in autonomous vehicles, facilitating enhanced coordination and decision-making in real-world driving scenarios.

DATA SCIENTIST

AXIO

July 2023 - July 2024

Bengaluru, India

Implemented Reject Inferencing and revamped Decision Engines using income and bureau features derived scores with Gradient Boosting Decision Trees reducing risk by 30% on Monthly Disbursals close to 12 million USD
Designed and implemented real-time data pipeline to process 2+ million raw records from 35+ sources for buy now pay later portfolio
Contributed to the formulation of models for fraud identification, controlling risk by 40% while minimizing disbursal losses to 20%
Optimized CI/CD workflows, achieving 45% more efficiency in model testing and deployment for Axio App with more than 500,000 users
Conducted A/B testing and experimental analysis to evaluate the impact of underwriting model on overall portfolio performance

INTERN

BAIN & COMPANY

January 2023 - June 2023

New Delhi, India

Developed and deployed models for tasks like sentimental analysis and customer segmentation in domains across FMCG, Manufacturing and Healthcare for clients in American, Middle Eastern and European markets
Constructed an end-to-end web application for insight generation, and directly making visualisations from secondary research by natural language interaction with RestAPIs for a custom knowledge graph-based RAG engine boosting productivity by 40%
Developed robust data analysis and visualization products to derive insights increasing KPI tracking efficiency by more than 50%
Directly engaged with clients while maintaining NPS score of perfect 10 during tenure for the team and converted Full-Time Role

STUDENT RESEARCH SCHOLAR - AI/ML

UNIVERSITY OF SOUTH CAROLINA (AIISC)

September 2022 - May 2023

Remote

Assisted Dr. Amit Seth and Dr. Amitava Das at the Artificial Intelligence Institute – University of South Carolina (AIISC)
Formulated a framework RESTORE, for analysis of intrinsic information retention of Graph embedding methods in preserving the semantic information on 1-3 hop reconstruction across 5+ Knowledge Graph datasets containing 2,000,000 + nodes and edges
Investigated and contributed in devising a novel reconstruction method CONFLUENCER, leveraging custom autoencoders with loss function combinations and 3 meta embedding generation methods for graphs using multiple sub graph embeddings

VISITING RESEARCH SCHOLAR

NORTHEASTERN UNIVERSITY

June 2022 - August 2022

Boston MA, USA

Designed a high-resolution temperature sensor and EEG data acquisition system under the guidance of NSF Award recipient Dr. Aatmesh Shrivastava for investigating correlation between neurological activity and minute body surface temperature change
Designed the circuit after multiple simulations LTSpice and the device was designed with Altium generating Gerber files for fabrication

UNDERGRADUATE RESEARCH ASSISTANT

BIRLA INSTITUTE OF TECHNOLOGY - MESRA

July 2021 - May 2022

Developed a novel deep learning-based Edge device for plant disease detection leveraging chromatic indices for added colorimetric feature extraction trained on PlanVillage Dataset achieving 97% accuracy with the on-device quantized model leading to a paper currently under review at Computers and electronics in agriculture- Elsevier
Proposed a hybrid architecture for retinal disease detection using Transformers based architecture in combination with machine learning techniques, achieving 94% accuracy with perfect precision and recall across 2 of 4 classes, presented at the 6th IEEE International Conference on Computing Methodologies and Communication (ICCMC 2022)
Leveraged ensemble learning techniques to detect faults and classify Power transmission lines achieving 98.6% accuracy which was presented at the 7th International Conference on Nano-electronics, Circuits & Communication Systems (NCCS-2021) and was also awarded Best Paper Award

Publications

My research papers and academic contributions

Deepfake Detection over Network Communication Channels

Jha, S., Nag, S., Liu, Z., Kundu, R., Bachu, S., Niloy, F. F., Tan, Z., Qian, Z., Krishnamurthy, S., & Roy-Chowdhury, A.

Under Review at an A* venue

2026

This paper presents DeFND, a comprehensive benchmark dataset for evaluating deepfake detection methods under network-induced degradations in real-world production environments.

Under Review

Crop Disease Classification using Support Vector Machines with Green Chromatic Coordinate (GCC) and Attention based feature extraction for IoT based Smart Agricultural Applications

Jha, S., Luhach, V., Gupta, G. S., & Singh, B.

Preprint

2023

This research explores innovative approaches to crop disease classification using machine learning techniques with specialized feature extraction methods for smart agricultural applications.

Preprint

RESTORE: Graph Embedding Assessment Through Reconstruction

Yip, H. Y., Ravuru, C., Banerjee, N., Jha, S., Sheth, A., Chadha, A., & Das, A.

Preprint

2023

This paper presents RESTORE, a framework for analyzing the intrinsic information retention capabilities of graph embedding methods through reconstruction-based assessment.

Preprint

CONFLUENCER: Automated Graph Merging through Vector-to-Graph Reconstruction with Autoencoders

Yip, H. Y., Ravuru, C., Banerjee, N., Jha, S., Sheth, A., Chadha, A., & Das, A.

Preprint

2023

This research introduces CONFLUENCER, a novel approach for automated graph merging that leverages vector-to-graph reconstruction techniques with autoencoder architectures.

Under Review @ A* Venue

Indian Peak Power demand Forecasting: Transformer Based Implementation of Temporal Architecture

Jha, S., Luhach, V.

IEEE Global Conference on Computing, Power and Communication Technologies (GlobConPT)

2022

This publication presents a transformer-based temporal architecture for forecasting peak power demand in India, providing insights for energy management and infrastructure planning.

Paper

Retinal Malady Classification using AI: A novel ViT-SVM combination architecture

Jha, S., Luhach, V., Poddar, R.

6th IEEE International Conference on Computing Methodologies and Communication (ICCMC)

March 2022

This research proposes a novel architecture combining Vision Transformers (ViT) based feature extraction with Support Vector Machines (SVM) classification for accurate detection of retinal diseases

Paper

Power transmission line classification through Artificial Intelligence: Application of Transfer Learning and Deep Convolutional Learning Techniques

Jha, S., Luhach, V., Nath, V.

7th International Conference on Nano-electronics, Circuits & Communication Systems (NCCS)

2021

This paper demonstrates the application of ensemble learning for power transmission line classification, achieving 98.6% accuracy in fault detection and classification.

Best Paper Award

Handbook of Bi-Polar Stepper Motor Control

Notion Press

Book Publication

A comprehensive handbook on bi-polar stepper motor control mechanisms and applications. One of the Acknowledged contributors of the book

ISBN: 978-1685231569

Projects

A selection of my most significant technical projects in AI, Computer Vision, and Safety

💎Gemstone Price Prediction MLOps Pipeline

Designed and implemented a complete MLOps pipeline for gemstone price prediction, featuring automated model training, deployment and monitoring. The system includes continuous integration, model versioning, and performance monitoring in production environments.

Technologies Used:

Docker AWS MLflow CI/CD Python

Code

🛫 US Immigration Data Engineering Pipeline

Developed a data engineering pipeline to process and analyze US immigration data, integrating with airport information and US demographics. The system employs distributed processing with Apache Spark, orchestration with Airflow, and cloud-based data warehousing with Amazon Redshift.

Technologies Used:

Apache Spark Apache Airflow AWS Redshift AWS S3 Python SQL

Code

🛡️ Toxicity Barrier Framework

Developed a robust multi-modal safety filtering system inspired by Control Barrier Functions. This framework provides real-time toxicity detection for text and images, with an adaptive safety constraint mechanism that ensures responsible content interaction across modalities.

Technologies Used:

Python PyTorch CLIP GPT-4 Vision Transformers

Code

🤖 Transformers from Scratch

A comprehensive implementation of transformer models built from the ground up using JAX and Flax. This educational project demonstrates the progressive development of transformer architecture, from simple bigram models to complex GPT-2 style implementations, providing deep insights into large language models.

Technologies Used:

JAX Python NLP Deep Learning

Code

Uncertainty Quantification in Occluded 3D Humans

Ongoing research focusing on robust uncertainty quantification for 3D human pose estimation under challenging occlusion scenarios. This work aims to improve reliability in human-computer interaction systems by providing confidence metrics for pose predictions.

Technologies Used:

PyTorch 3D Computer Vision Bayesian Deep Learning OpenCV

Ongoing

Multi-Agent Autonomous Vehicle Perception

Developing a late fusion approach for multi-agent perception in autonomous vehicle systems. This research explores collaborative sensing across multiple vehicles to improve overall perception accuracy, reliability, and range in complex driving scenarios.

Technologies Used:

TensorFlow LiDAR Sensor Fusion V2X Communication ROS

Ongoing

🔍 Enhanced Vision-Language-Action Models

Ongoing research focused on enhancing the capabilities of foundational vision-language-action models by improving context integration in the embedding pipeline. This work aims to create more robust multimodal systems that better understand the relationships between visual inputs, language instructions, and action planning.

Technologies Used:

PyTorch Transformers CLIP Multimodal Learning Embodied AI

Ongoing

🧠 Neural Networks from Scratch in C++

Implemented a comprehensive neural network library in C++ without relying on deep learning frameworks. This project includes implementations of various neural network architectures, activation functions, optimization algorithms, and backpropagation from first principles, providing deep insights into the mathematical foundations of deep learning.

Technologies Used:

C++ Linear Algebra CUDA

Code

Volunteering & Leadership

My commitment to community service and leadership experiences

Leadership

Founding President

180 Degrees Consulting BIT Mesra

2021 - 2023

Established and led the BIT Mesra branch of 180 Degrees Consulting, the world's largest student-run non-profit consulting organization.

Key Achievements:

Led the team to be among only 3 Indian universities in 2021 selected by 180DC to set up a branch through the global accelerator program
Successfully completed 5 consulting projects for non-profit organizations during the first year of establishment
Organized and conducted 6 professional development workshops for student consultants

Pulpit Co-ordinator

UNESQUO (The quizzing and debating society) BIT Mesra

2021 - 2022

Coordinated quizzing and debating activities for the official quizzing and debating society of BIT Mesra.

Key Responsibilities:

Organized quizzing and debating competitions at the institutional level
Facilitated training sessions for new members to develop public speaking and debate skills
Represented the society in inter-college competitions and events

Joint Secretary

EDC (Entrepreneurship Development Cell) BIT Mesra

2021 - 2022

Served as a key member of the administrative team for the Entrepreneurship Development Cell at BIT Mesra.

Key Responsibilities:

Assisted in planning and executing entrepreneurship-focused events and workshops
Managed communication between the cell and various stakeholders, including students, faculty, and external partners
Contributed to initiatives aimed at fostering an entrepreneurial mindset among students

Volunteering

MENTOR

Desh Ke Mentor

January 2022 - June 2022

Volunteered as a Mentor in the nation's largest mentoring program launched by the Delhi government under the guidance and leadership of CM Arvind Kejriwal and Dy CM Manish Sisodia.

Impact:

Mentored 5 students between ages 15 to 18 years, providing guidance on various aspects of their development
Provided counseling regarding employment opportunities, higher education options, and career pathways
Supported mentees with guidance on mental and physical well-being

Shashwat Jha

Skills

LANGUAGES

TOOLS & FRAMEWORKS

LIBRARIES

DOMAINS

Experience

ASSOCIATE AI ENGINEER

STUDENT RESEARCHER

DATA SCIENTIST

INTERN

STUDENT RESEARCH SCHOLAR - AI/ML

VISITING RESEARCH SCHOLAR

UNDERGRADUATE RESEARCH ASSISTANT

Publications

Deepfake Detection over Network Communication Channels

Crop Disease Classification using Support Vector Machines with Green Chromatic Coordinate (GCC) and Attention based feature extraction for IoT based Smart Agricultural Applications

RESTORE: Graph Embedding Assessment Through Reconstruction

CONFLUENCER: Automated Graph Merging through Vector-to-Graph Reconstruction with Autoencoders

Indian Peak Power demand Forecasting: Transformer Based Implementation of Temporal Architecture

Retinal Malady Classification using AI: A novel ViT-SVM combination architecture

Power transmission line classification through Artificial Intelligence: Application of Transfer Learning and Deep Convolutional Learning Techniques

Handbook of Bi-Polar Stepper Motor Control

Projects

💎Gemstone Price Prediction MLOps Pipeline

Technologies Used:

🛫 US Immigration Data Engineering Pipeline

Technologies Used:

🛡️ Toxicity Barrier Framework

Technologies Used:

🤖 Transformers from Scratch

Technologies Used:

Uncertainty Quantification in Occluded 3D Humans

Technologies Used:

Multi-Agent Autonomous Vehicle Perception

Technologies Used:

🔍 Enhanced Vision-Language-Action Models

Technologies Used:

🧠 Neural Networks from Scratch in C++

Technologies Used:

Volunteering & Leadership

Leadership

Founding President

Key Achievements:

Pulpit Co-ordinator

Key Responsibilities:

Joint Secretary

Key Responsibilities:

Volunteering

MENTOR

Impact:

Resume

Collaborations & Affiliations

Contact Me